A method of constructing a confusion set database for use in detecting user query intentions includes obtaining a bilingual database having aligned word pairs in first and second languages. Second language word pairs in the bilingual database are aligned with corresponding correct translation first language word pairs. First language human translation word pairs corresponding to each of the second language word pairs in the bilingual database are obtained. Each first language human translation word pair for a particular second language word pair in the bilingual database is aligned with the correct translation first language word pair to define first language set pairs in the confusion set database. Methods, systems and computer readable medium for constructing the confusion set database and for retrieving sentences using the confusion set database are also disclosed.